Google have realeased a paper Scalable and accurate deep learning for electronic health records：https://arxiv.org/pdf/1801.07860.pdf which arouse so many discussion here in China. what can we/OHDSI community can learn from this single study and anyone know where can i get more similar research based OHDSI?
such as Inpatient mortality, 30-day unplanned readmission, Long length of stay ,Diagnoses
1 Like
@wanghaisheng
I’ve been developing scalable RNN model based on CDM, which was presented as a poster in OHDSI 2017 (code).
As a pilot study, I predicted long-term mortality by using claim data. I planned to apply this model on hospital CDM to predict in-hospital mortality.
This model will be integrated into PLP package with help of @Rijnbeek github .
You can find this model one or two months later in PLP package.
Hi Haisheng,
Yes, that is an interesting paper, we are also working on PatientLevelPrediction modeling in OHDSI, including deep learning, lead by @Rijnbeek in Erasmus MC, Netherlands, maybe we can work together on some topics:-)
By the way, we are also organizing OHDSI European Symposium https://www.ohdsi.org/events/ohdsi-europe-symposium/, if you know someone who maybe interest in china, please help distribute, thanks.
I also came across that paper and found it interesting. I am new to OHDSI and to this area (first year PhD student) but would love to become further involved in this area in any way I can. Please let me know if anything comes up that you could use some help with!
In this paper, they said that
‘It is widely held that 80% of the effort in an analytic model is preprocessing, merging, customizing, and cleaning data sets, not analyzing them for insights. This profoundly limits the scalability of predictive models.’
I think CDM based predictive algorithm (Thanks to @Rijnbeek, @jennareps, @xypan1232) will be very helpful for scalability of predictive models.
1 Like